Data Exfiltration via Prompt Injections in Copilot Studio Agent

# Data Exfiltration via Prompt Injections in Copilot Studio Agent

https://gcdradar.vercel.app/incidents/data-exfiltration-via-prompt-injections-in-copilot-studio-ag · AIR-2025-0059 · closed 2025-11-21

- Lane: Research
- Severity: 1 (Research / proof of concept)
- Fix status: Fix status unknown
- First reported: 2025-11-07 (happened 2025-06)
- AI role: Attack on AI
- Products: Copilot Studio

## Summary

Researchers demonstrated how an organization's data could be exfiltrated through prompt injections targeting an AI customer service agent. The attack exploited tool access in a Copilot Studio implementation.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2025-11-07: AgentFlayer: When AIjacking Leads to Full Data Exfiltration in Copilot Studio (https://labs.zenity.io/p/a-copilot-studio-story-2-when-aijacking-leads-to-full-data-exfiltration-bc4a)
- 2025-11-07: AgentFlayer: Discovery Phase of AI Agents in Copilot Studio (https://labs.zenity.io/p/a-copilot-studio-story-discovery-phase-in-ai-agents-f917)
- 2025-11-07: Data Exfiltration via Agent Tools in Copilot Studio (https://atlas.mitre.org/studies/AML.CS0037)

## Sources

- 2025-11-07 MITRE ATLAS: Data Exfiltration via Agent Tools in Copilot Studio https://atlas.mitre.org/studies/AML.CS0037
- 2025-11-07 Zenity Labs: AgentFlayer: Discovery Phase of AI Agents in Copilot Studio https://labs.zenity.io/p/a-copilot-studio-story-discovery-phase-in-ai-agents-f917
- 2025-11-07 Zenity Labs: AgentFlayer: When AIjacking Leads to Full Data Exfiltration in Copilot Studio https://labs.zenity.io/p/a-copilot-studio-story-2-when-aijacking-leads-to-full-data-exfiltration-bc4a

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.